Custom Query (245 matches)

Porting cdrecord, I found very strange effect, when using printf function developed by cdrecord author (Jorg Schiling). Futher analisys brought this result.

I provide a testcase. Please compare results with and without -O2 compiler switch.

The EMXSHELL is used to override COMSPEC for popen() and system() calls in kLIBC. The order of checking in the first case is EMXSHELL, COMSPEC and in the second — EMXSHELL, COMSPEC, OS2_SHELL, SHELL.

At the same time, many scripts in the unix environment (e.g. configure scripts) use SHELL to override the shell command to use. However, given that in the first case SHELL isn't used at all and in the second case it's checked after EMXSHELL and COMSPEC, even if the script sets SHELL (e.g. to point to sh.exe), kLIBC will not notice this and will still use the value of COMSPEC (which is usually cmd.exe). Not quite what the scripts expect.

I think it makes sense to check for SHELL before checking for EMXSHELL or COMSPEC. The order should be the same in both calls: SHELL, EMXSHELL, COMSPEC, OS2_SHELL.

The patch is trivial (in files lib/io/popen.c and lib/process/system.c) so not attaching it here.

Output buffers are not flushed if file handle is not closed before program end. Sample testcase:

#include <stdio.h>
#include <string.h>

void main(void)
{
	char* buf = "this is my test buffer";

	FILE *fo = fopen( "test", "w");
	fwrite( buf, 1, strlen( buf), fo);
}

leaves a 0-byte file on disk. Adding flush or fclose restores proper behaviour.